A single gradient step finds adversarial examples on random two-layers neural networks.
Sébastien BubeckYeshwanth CherapanamjeriGauthier GidelRemi Tachet des CombesPublished in: NeurIPS (2021)
Keyphrases
- neural network
- multi layer
- pattern recognition
- genetic algorithm
- single layer
- fuzzy logic
- recurrent neural networks
- feed forward
- self organizing maps
- back propagation
- post processing
- edge detection
- multi agent
- multilayer perceptron
- computational intelligence
- neural nets
- training algorithm
- small size
- uniformly distributed
- single step
- information systems
- multiple layers